The Generality of Deviance in Late Adolescence and Early Adulthood D. Wayne Osgood; Lloyd D. Johnston; Patrick M.

O'Malley; Jerald G. Bachman American Sociological Review, Vol. 53, No. 1. (Feb., 1988), pp. 81-93.
Stable URL: http://links.jstor.org/sici?sici=0003-1224%28198802%2953%3A1%3C81%3ATGODIL%3E2.0.CO%3B2-R American Sociological Review is currently published by American Sociological Association.

Your use of the JSTOR archive indicates your acceptance of JSTOR's Terms and Conditions of Use, available at http://www.jstor.org/about/terms.html. JSTOR's Terms and Conditions of Use provides, in part, that unless you have obtained prior permission, you may not download an entire issue of a journal or multiple copies of articles, and you may use content in the JSTOR archive only for your personal, non-commercial use. Please contact the publisher regarding any further use of this work. Publisher contact information may be obtained at http://www.jstor.org/journals/asa.html. Each copy of any part of a JSTOR transmission must contain the same copyright notice that appears on the screen or printed page of such transmission.

The JSTOR Archive is a trusted digital repository providing for long-term preservation and access to leading academic journals and scholarly literature from around the world. The Archive is supported by libraries, scholarly societies, publishers, and foundations. It is an initiative of JSTOR, a not-for-profit organization with a mission to help the scholarly community take advantage of advances in technology. For more information regarding JSTOR, please contact support@jstor.org.

http://www.jstor.org Thu Nov 29 03:29:19 2007

THE GENERALITY OF DEVIANCE

IN LATE ADOLESCENCE AND EARLY ADULTHOOD*

D. WAYNE OSGOOD
University of Nebraska, Lincoln

LLOYD JOHNSTON, D. M PATRICK . O'MALLEY, JERALD . BACHMAN G

The University of Michigan

Because a wide variety of deviant behaviors are positively correlated with one another, some researchers conclude that all are manifestations of a single general tendency. The present analysis incorporated three waves of self-reports about heavy alcohol use, marijuana use, use of other illicit drugs, dangerous driving, and other criminal behavior for a nationally representative sample of high school seniors. A relatively stable general involvement in deviance accounted for virtually all association between different types of deviance, but the stability of each behavior could only be explained by equally important and stable specific influences. Thus, theories that treat different deviant behaviors as alternative manifestations of a single general tendency can account for some, but far from all, of the meaningful variance in these behaviors. The only significant influence of one type of deviance on another was that of marijuana use on later use of other illicit drugs. The causal model also revealed interpretable shifts zn the associations among these behaviors over the four years following high school.

INTRODUCTION Research has firmly established that a wide range of deviant behaviors are positively correlated with one another during adolescence and early adulthood (e.g., Akers 1984; Donovan and Jessor 1985; Elliott and Huizinga 1984; Johnston, O'Malley, and Eveland 1978). This paper concerns the sources of that association and its significance for theories of deviance. We will be particularly concerned with the possibility that deviance is a unified phenomenon, with various behaviors serving as alternative manifestations of a more general tendency. There are two plausible general explanations for correlations among deviant behaviors. The first is that engaging in one form of deviant behavior leads to engaging in others as well. Many people believe that there are causal links between some forms of deviance, particularly that drug use leads to crime. The second explanation is that different deviant behaviors are related because they have shared influences. For example, the factors that lead people to become sexually active at an early age might be the same as (or at least overlap) those that lead them to use marijuana. To the degree that the same factors are major sources of all deviant

* Direct D. Osgood, Department of Sociology, University of Nebraska, Lincoln. NE 68588-0324. This research was supported by grant number DA 01411 from the National Institute on Drue Abuse. The authors thank Ron Kessler and Jon ~ r o s n z k helpful for discussions of the analysis.

behaviors, it is meaningful to speak of a general syndrome of deviance (Donovan and Jessor 1985). Sociologists have offered many definitions of deviance. (For a general discussion, see Gibbs 1981, ch. 2.) Our concerns center on behavior socially defined as undesirable rather than on the social processes that lead certain individuals to be labeled deviants. Jessor and Jessor (1977) offer a clear definition of deviance (which they refer to as problem behavior): it is "behavior that is socially defined as a problem, a source of concern, or as undesirable by the norms of conventional society and the institutions of adult authority, and its occurrence usually elicits some kind of social control response." ( p . 33) Our study examines several deviant behaviors. By definition, all deviant behaviors violate conventional standards of behavior. Even so, each deviant behavior may be a unique phenomenon requiring a separate explanation, or the various deviant behaviors may form a unified phenomenon with a single explanation. This is an empirical question with major import to theories of deviance. The generality of deviance across different types of behavior will be a function of the degree to which the behaviors have the same influences. Generality versus specificity is a relevant issue in many areas o f ~ s o c i o 1 0 ~ ~ . instance, For we speak of social status as encompassing income, education, and occupational prestige; the transiency, pove*y9 and physical deterioration of a community are all signs of social disorganization; and social stratification includes income inequality, residential segrega81

American Sociological Review, 1988, Vol. 53 (February:81-93)

AMERICAN SOCIOLOGICAL REVIEW tion, and social mobility. These are meaningful groupings that have conceptual coherence, but they may or may not constitute unified empirical phenomena. If social stratification is empirically general across its different manifestations, then explaining one form of stratification is sufficient to explain them all. In this case, plausible theories of stratification would look quite different than if the different forms occur independently of one another. Our framework for assessing the generality of deviance is applicable to other concepts as well. Our approach to analyzing generality versus specificity transcends any particular theoretical position. We do not focus on the role of an a priori set of explanatory variables, such as Jackson et al.'s (1986) assessment of specificity based on differential association theory. Instead, we partition all reliable variance into general and specific components on the basis of covariance among various deviant behaviors in a longitudinal research design. SHARED INFLUENCES AND THE GENERALITY O F DEVIANCE Most sociological theories are consistent with deviance being general across different behaviors. Almost any explanation offered for one behavior has been offered for others as well. For instance, social scientists have argued that peer influence leads to early cigarette smoking (Krosnick and Judd 1982), early sexual intercourse (Billy and Udry 1985), marijuana use (Kandel 1978), and criminal behavior (Sutherland and Cressey 1955). Models emphasizing social learning (Akers 1977), subcultural norms (Coleman 196 I), enhancement of self-esteem (Kaplan 1975), and social bonds (Hirschi 1969) have been proposed for a variety of deviant behaviors, and some theorists have simultaneously addressed several forms of deviance within a single explanatory framework (e.g., Elliott, Huizinga, and Ageton 1985; Jessor and Jessor 1977; and Kaplan 1975). While many social explanations indicate how different deviant behaviors might have influences in common, explanations vary in the degree to which the process causing one deviant behavior will jointly produce others. For instance, Akers' social learning approach (1977) explains each deviant behavior as a result of associating with people who model and reinforce that behavior. The processes of modeling and reinforcement lead to both alcohol use and theft only to the degree that associates who support one also support the other. Thus, in Akers' theory the generality of deviance is dependent on the empirical correlation between conceptually independent variables. The strongest position in favor of the generality of deviance is that different deviant behaviors are manifestations of a single underlying construct. Jessor and his colleagues have posited that a variety of deviant behaviors form a "syndrome," which is directly caused by a general latent variable of unconventionality (Donovan and Jessor 1985; Jessor and Jessor 1977). Hirschi (1984) explains the relationship between drug use and delinquency in a similar manner, stating that the two are not merely influenced by some of the same factors, but "they are manifestations of the same thing" (p. 51). This "thing" is criminality, which he defines as "the tendency or propensity of the individual to seek short term, immediate pleasure" (p. 51). According to Hirschi's social control theory (1969), criminality results from the absence of social bonds. Hirschi and Gottfredson have recently articulated wideranging theoretical ramifications of the concept of criminality, with its image of deviant behavior as a manifestation of general and relatively stable individual differences (Gottfredson and Hirschi 1986; Hirschi and Gottfredson 1983, 1986). For our purposes, the most important implication of the positions taken by Jessor and colleagues and by Hirschi and Gottfredson is that explaining a general tendency toward deviance is sufficient to account for a large group of behaviors and that causes specific to any particular form of deviance are relatively unimportant. The intermediate position is that a general cause, alienation from the norms of conventional society, is a partial determinant of a range of deviant behaviors. Cloward and Ohlin (1960) argue that the community's illegitimate opportunity structure shapes the specific deviance that will result, while Elliott et al. (1985) point to social learning from the peer group. Such theories predict substantial, but not complete, generality of deviance. INFLUENCE O F ONE DEVIANT BEHAVIOR ON ANOTHER Taking the notion of the generality of deviance to its limit would preclude any influence of one type of deviant behavior on another. In this case, influences specific to particular forms of deviance not only would be unimportant, but nonexistent. Given a propensity toward deviance, the specific deviant behaviors in which a person engages at any time would be strictly random. It then follows that there would be perfect correlations among different deviant behaviors, limited only by the reliability of their measures. Since there would be nothing unique about specific behaviors, they could not possibly influence one another. Short of this extreme position, however, it is

GENERALITY OF DEVIANCE possible for different deviant behaviors to have some influences in common and some influences that are specific, including one behavior serving as a partial cause of another. Though we know of no theories that explicitly predict influences between specific deviant behaviors, such influences would be consistent with several theories.1 Consider a person who begins to use marijuana regularly. This behavior could lead to rejection by conventional peer groups and increased association with peer groups that approve of various forms of deviance. If so, social learning from the new group could result in other deviant behaviors, as could weakened bonds to conventional groups. If the marijuana use was detected by authority figures, then labeling theory predicts the development of a deviant identity and secondary deviance (Lemert 1972), which presumably would encompass a variety of behaviors. determining the importance of shared versus specific influences on different deviant behaviors. Whatever the findings, their implications are limited to the role of the finite set of explanatory variables included in the research. The shared influences identified by Elliott et al. and by Jessor and Jessor might be of little consequence compared to specific influences from variables they did not study. Conversely, the specific influences identified by Kandel et al. and Johnston might be trivial departures from a larger pattern of shared influences. Directly assessing the generality of deviance requires an approach that is not limited to measured influences. The work of Donovan and Jessor (1985) illustrates such an approach. They focused on covariance among deviant behaviors rather than on the relationships of the behaviors to potential explanatory variables. Using confirmatory factor analysis, Donovan and Jessor determined that a variety of behaviors formed a general syndrome of deviance. Their results indicate that a single latent variable is sufficient to account for covariance among the behaviors, and they replicate this result for several samples. Nevertheless, further evidence is needed before accepting Donovan and Jessor's conclusions. Their analyses are cross-sectional, and, therefore, they could not distinguish between covariance due to shared influences and covariance due to an influence of one behavior on another. Furthermore, not all behaviors were well explained by the general syndrome. In many instances, the latent variable accounted for less than 10 percent of the variance of specific behaviors. Though their conclusions imply that any remaining variance was simply error of measurement, their method provides no means of differentiating error of measurement from meaningful variance specific to a particular variable. Two studies have used longitudinal data from nationally representative samples to assess influence between crime (or delinquency) and illicit drug use. Both Johnston, O'Malley, and Eveland (1978) and Elliott and Huizinga (1984) concluded that shared influences are the major source of the relationship between these behaviors and that influence of one behavior on another is relatively unimportant. Nevertheless, neither study strictly rules out the possibility of influence between behaviors, and each provides some evidence of such influences. Johnston et al.'s cross-lag panel analysis yielded path coefficients consistent with modest reciprocal influence between the behaviors (p. 151). Similarly, Elliott and Huizinga found that earlier delinquency predicted later drug use, even after taking earlier drug use into account (p. 88). The present study uses modem techniques of

Available Evidence
Several studies are pertinent to the generality of deviance and the sources of positive correlations between various deviant behaviors. Findings about shared influences appear in studies that relate the same explanatory variables to several different deviant behaviors and in analyses of the factor structure of covariance among behaviors. Influence between specific deviant behaviors has been investigated through longitudinal research measuring the same behaviors at two or more times. Elliott et al. (1985) and Jessor and Jessor (1977) investigated a variety of causal factors, and their findings support the possibility that shared influences create relationships between different deviant behaviors. Factors that strongly influenced one deviant behavior (such as delinquency) similarly influenced other behaviors (such as alcohol and drug use). On the other hand, some research indicates that certain causal factors are more important for one deviant behavior than for others. For instance, Kandel, Kessler, and Margulies (1978) concluded that parental influences were much less important for marijuana use than for use of other illicit drugs. Johnston (1973) found that ideological alienation related to some forms of illicit drug use, but not others, and not to cigarette use, alcohol use, or delinquency. These findings imply that involvement in deviance is not completely general across behaviors. Studies such as these have limited value for
The possibility that physical need due to drug addiction leads to increased crime falls outside our interest in social explanations of deviance, and physical addiction is rare in our sample.

84

AMERICAN SOCIOLOGICAL REVIEW

Morijuono
Use

T~me One

Time Two

T~me Three

Fig. 1. A Causal Model Differentiating General and Specific Influences on Deviant Behaviors

causal modeling for panel data (e.g., Kessler and Greenberg 198 1) to advance our understanding of the relationships between different forms of deviance. We develop a structural equation model that separates general and specific components of each behavior. The longitudinal aspect of our study allows us to examine the importance of general involvement in deviance not only in terms of the size of the general component for each behavior (as in Donovan and Jessor's 1985 analysis), but also to compare it directly to reliable specific variance and to assess the stability of both over time. Our test for influence between specific variables improves on earlier work by addressing several types of deviance, by explicitly modeling shared influences, and by allowing for error of measurement. Finally, by covering several years of the age span, the model can relate changing age norms to shifts in the strength of connections between individual behaviors and a general syndrome of deviance. METHOD

specific to each behavior, and influence of one behavior on another. The causal model is unusual in that each behavior serves as an indicator of the general factor and also has unique aspects that may influence other variables. In effect, we divide measures of behavior into three components: variance shared with a general tendency toward deviance; reliable variance specific to the behavior; and error of measurement. The model includes disturbance terms for all latent variables as well. The model will be identified given three conditions. First, there can be no correlated errors between the general and specific latent variables. Second, there must be constraints on the error terms of the observed variables, or the specific variances will be undefined. We obtained meaningful estimates by assuming that all influence of wave one on wave three is mediated by wave two and that reliability is constant over time2 (Kessler and Greenberg

Causal Models Our basic causal model is illustrated in Figure 1. For simplicity of presentation, this figure is limited to two behaviors: criminal behavior and marijuana use. The model incorporates a general tendency toward deviance as a contemporaneous influence on all of the behaviors. It allows for longitudinal influences in the form of stability of that general tendency, stability of variance

' A n equally plausible assumption is that error variance, rather than reliability, was constant. For these data, the assumption of equal reliability proved superior: it yielded consistent and interpretable estimates, while the assumption of constant error variance resulted in negative estimates of residual variance. Because reliability refers to the proportion of error variance (rather than the absolute amount of error variance), it was necessary to estimate the model from the correlation matrix. The presented coefficients have been transformed to reflect the original metric of the variables (as in an analysis of a covariance matrix).

GENERALITY O F DEVIANCE 1981, p. 147-50). Third, the number of paths among the specific variances must be limited. Within each wave, one degree of freedom per behavior is absorbed by estimating influences from general deviance, and one degree of freedom between adjacent waves is absorbed by estimating the stability of general deviance. The former reduces the number of possible error paths among contemporaneous specific variances, while the later reduces the number of possible longitudinal influences between them. The importance of the general tendency is reflected both by the strength of its paths to each behavior and by its stability over time. If all the behaviors are interchangeable manifestations of the general tendency (the extreme version of the shared-influence hypothesis), there will be no meaningful variance specific to the separate behaviors. In this case, a simpler model would be adequate to account for the data, a model in which each behavior is comprised only of general deviance and error of measurement, and the stability of general deviance is the' only longitudinal influence. On the other hand, the presence of a substantial amount of stable specific variance would indicate that the various forms of deviance cannot be fully explained by a single causal process. Because the model in Figure 1 divides each deviant behavior into general and specific components, it separates the influence of one behavior on another from their associations with general deviance. Influences between behaviors are indicated by paths from one specific variance to a different specific variance at the next time. The dashed lines in Figure 1 represent such an influence of marijuana use on criminal behavior. It is important that the model also take into account error of measurement; failing to do so would cause bias in estimates of causal paths. Given a set of uniformly positively related variables such as these, the bias will be toward spurious positive influences among behaviors. This might explain the apparent influence between drug use and delinquency in earlier research (Elliott and Huizinga 1984; Johnston et al. 1978).
Sample

in 1975, a wide range of information is gathered from a nationally representative sample of high school seniors each year.' Data for the present analysis come from the follow-up portion of the study, which is based on a subsample of each senior class. Half of the participants in the follow-up study complete questionnaires in every odd-numbered year after graduation, and the other half do so in every even-numbered year. The present analysis used three waves of data, provided at approximate ages of 18, 19, and 21, or 18, 20, and 22. The analysis was limited to white respondents, since the black subsample is somewhat less representative due to differential high school drop-out rates. The follow-up study over-samples the more serious drug users in high school to obtain more accurate estimates for this segment of the population. The over-sampled individuals are then given smaller weights in analyses to produce a representative sample. There were 975 respondents in the sample, coming from the high school senior classes of 1976 though 1980, yielding a weighted sample of 717 cases. Initial involvement in the behaviors being studied typically occurs at earlier ages than are included in this sample (e.g., Elliott and Huizinga 1984). We do not consider this a shortcoming of our study. We reject the point of view that a deviant behavior is "caused" at some time of "onset," after which it is self-perpetuating until it is "caused" to stop. Instead, we assume that a behavior occurs when its causes are present and does not occur when they are absent. As our results demonstrate, the stability of these behaviors is far from perfect, meaning that there is a great deal of change independent from the general age trends. Thus, there is just as much need to explain persistence of deviant behavior as to explain its onset and cessation. As Hirschi (1984, p. 50) has articulated, it is not at all clear that "onset" is a meaningful concept for deviant behavior. It is only in retrospect that getting drunk for the first time can be called the onset of alcoholism. Asking whether the first marijuana cigarette precedes the first incident of theft is a much less meaningful question than asking whether current marijuana use has an influence on later criminal behavior.
"ach year, a three-stage national probability sample leads ro questionnaire administrations in approximately 130 high schools (roughly 110 public and 20 private). This procedure yields between 15.000 and 19,000 respondents. A random one-fifth of each annual sample completes the version of the questionnaire that includes the items used in the present analysis. Response rates for the base year average 80 percent. and follow-up response rates are generally 75 percent or more of the original group.

The data we analyze were collected as part of the Monitoring the Future study (Osgood, Johnston, O'Malley, and Bachman 1985). For a detailed description of the sample design and data collection, see Bachman and Johnston (1978). For a full listing of variables, see Johnston, Bachman, and O'Malley (1986). For detailed findings on drug use and related variables, see Johnston, O'Malley, and Bachman (1986). In this ongoing study, which began

86
Table 1. Goodness of Fit of Alternative Models

AMERICAN SOCIOLOGICAL REVIEW

I. Null Model Measurement models 11. All possible within-wave relationships, no longitudinal relationships 111. One general factor at each wave, no correlated errors, no longitudinal relationships IV. One general factor at each wave, 11 correlated errors, no longitudinal relationships Longitudinal models V. Stable general factor, no stable specific factors VI. Stable general and specific factors VII. Stable general and specific factors, plus longitudinal influence among all specific factors VIII. Stable general and specific factors, plus influence of marijuana use on other illicit drug use
Note: Models V-VIII include 10 correlated errors (see footnote 5).

A ,000
,461 ,427 ,460 ,715 ,982 ,992 ,986

d.f. 105
75 90 79 78 63 43 61

x2
4,563.94 2,461.63 2,615.21 2,465.94 1,298.88 80.11 36.07 63.90

P 0. 0. 0. 0. 0. .07 .76 .38

Measures

Our analysis is based on self-report measures of five different types of deviant behavior: criminal behavior (limited to illegal behavior directed at victims); heavy alcohol use; marijuana use; use of other illicit drugs; and dangerous driving. We chose these behaviors because they represent a broad range of the conventionally proscribed activities that are common during this age span. Research on the factor structure of substance abuse has shown that use of alcohol, marijuana, and hard drugs are relatively distinct phenomena (e.g., Hays et a1 1986), so we consider it appropriate to treat them separately. Dangerous driving is not among the deviant behaviors typically studied by social scientists. Nevertheless, it is quite appropriate to our definition of deviance since it is generally recognized as undesirable and is subject to social controls. Furthermore, it is an exciting, risky activity that follows the same age trend as criminal behavior (Hirschi and Gottfredson 1983). These five behaviors do not exhaust the concept of deviance, and our results may or may not characterize other types of deviance. The measure of heavy alcohol use referred to behavior in the past two weeks, and the other four measures referred to behavior over the last 12 months. The 14-item measure of criminal behavior also was used in the Youth in Transition study (Bachman, O'Malley, and Johnston 1978), and is an adaptation of Gold's (1970) well-known measure. These items concern interpersonal aggression, theft, and vandalism.4 Our index was a sum across the items,

each of which ranged from zero to four (for committing the act five or more times). Heavy alcohol use was measured in terms of the number of occasions a respondent had five or more drinks in a row during the last two weeks, with scores ranging from zero through five (10 or more). The scale for marijuana use varied from zero through nine (40 or more times in the last month). Use of other illicit drugs was measured as an average across eight drugs, each scored on the same scale as marijuana use. The measure of dangerous driving was the sum of reports of traffic tickets and traffic accidents, each scored as zero through four (four or more). RESULTS
Goodness of Fit of Alternative Models

We compared the goodness of fit of several alternative causal models to determine which explanatory factors account for observed relationships between the five deviant behaviors. The factors of interest were a general tendency toward deviance (representing shared influences); variance specific to a particular behavior (representing specific influences); and influences between particular behaviors. Table 1 summarizes the fit of the models, which were estimated by LISREL IV (Joreskog and Sorbom 1978). Chi-square values and their associated
actions that intentionally victimize other people. The other measures pertain to either victimless crimes or traffic.offenses (where victimization is rarely intended). For respondents who had not yet reached the age of majority, the variable technically refers to delinquency rather than crime. Some analyses of these items have distinguished interpersonal aggression from property offenses (e.g., Bachman et al. 1978). Preliminary analyses revealed negligible reliable variance in aggressive offenses at waves 2 and 3, meaning that separating these offenses would be fruitless for the present study.

The label "criminal behavior" is simply heuristic, since the other types of deviant behavior also may involve violating laws. The label should not be too misleading, however, since this aggregation is limited to

GENERALITY OF DEVIANCE probability levels indicate the lack of fit between the models and the observed covariances among the 15 variables (five deviant behaviors, each measured on three occasions). The probability levels should not be taken literally, however, since these skewed data do not justify the assumption of multivariate normality. A particularly useful index of fit is Bentler and Bonett's A (1980), which is independent of sample size. Model I is the "null model," which corresponds to the assertion that all of the variables are unrelated, and A is the proportionate reduction of this chi-square provided by other models. Measurement model. Our strategy in developing the causal model was to obtain an adequate within-wave measurement model before testing alternative longitudinal models. Our goal for the measurement model was to divide the reliable variance of each behavior into general and specific components. Cross-sectional covariance among the five behaviors provides a basis for defining a general factor of deviant behavior. Donovan and Jessor (1985) concluded that such a single latent variable was sufficient to account for cross-sectional relationships among a variety of deviant behaviors. Model I1 serves as a standard for the fit of the measurement model to the cross-sectional relationships, since it incorporated all possible relationships within each wave but did not allow any relationships between waves. The reduction in X 2 provided by this model is equal to the sum of the X 2 values for the null models of the three within-wave covariance matrices. For Model 11, A = ,461, meaning that 46 percent of the total chi-square value was due to cross-sectional relationships and 54 percent was due to longitudinal relationships. Model I11 is the basic measurement model, allowing for a single general factor at each wave. This model explained a large share of the cross-sectional relationships (A = .427), though a significant amount of within-wave covariation remained (comparing Models I1 and 111: AII.III = ,034, d.f. = 15, x = 153.58, p = 0.). This finding is in substantial agreement with the results of Donovan and Jessor (1985), in that a general factor accounts for 93 percent of the chi-square value attributable to within-wave relationships (AIII/AII).Though we found significant lack of fit for a single-factor model where they did not, this is likely due only to our larger sample size. The discrepancy in fit between Models I1 and I11 shows that some pairs of deviant behaviors were more strongly correlated with one another than is consistent with a single-factor model. By allowing some correlated errors between behaviors, we obtained more accurate estimates of the variances shared with the general factor. Furthermore, failing to incorporate such paths in other models could lead to spurious longitudinal paths. Model IV allowed for several correlated errors within each wave in addition to a general factor. This model accounted for virtually all of the within-wave covariance (AII.Iv = ,001, d.f. = 4, X 2 = 3.3 1, p > .25), and it served as the measurement model for the longitudinal models. It incorporated correlated errors between heavy drinking and marijuana use, marijuana use and other illicit drug use, criminal behavior and dangerous driving, and heavy drinking and dangerous d r i ~ i n g . ~ Longitudinal models. Model V was estimated to test the extreme version of the sharedinfluences hypothesis. In this model, all longitudinal relationships are explained by the stability of a general tendency toward deviance. The only latent variable at each wave was general deviance, so all specific variance was treated as error variance in the observed variables. Model V accounted for roughly half of the longitudinal covariation (Av = ,715; Av.II/[l AII] = .47), and left a highly significant and substantively important portion of the total covariance unexplained. Clearly, an adequate model requires longitudinal influences involving specific components. The remaining models are variations on the model illustrated in Figure 1. Each observed variable was modeled as a function of a latent variable of general deviance, a latent variable of specific variance, and error of measurement. These models included paths for the stability of the latent variables, and we assumed that all influence of wave one on wave three was mediated by wave two and that reliability was constant over time. Model VI expanded on Model V by allowing for a stable, specific component of each behavior as well as for general deviance; it did not allow for any influences between different forms of deviance. This model fit the data quite well (A = .982), indicating that the vast majority of longitudinal covariation was attributable to the stabilities of the general factor and of the reliable variance specific to each behavior.

To avoid inflated estimates of the variance each measure shared with the common factor, only positive correlated error paths were included. The path between heavy drinking and dangerous driving became negative and insignificant for the third wave, so it was eliminated. In models incorporating longitudinal influences (Models V-VIII), the path between marijuana use and other illicit drug use was removed for the same reason. The correlated error paths were between the latent variables in Models VI-VIII and between the observed variables in Models IV and V (which did not have latent variables for specific variance).

AMERICAN SOCIOLOGICAL REVIEW

Table 2. Variance Components and Reliabilities Variance Components Mean Variance Rel. General Specific Error Proportion of Reliable Var. General Specific

Criminal behavior Time 1 Time 2 Time 3 Heavy alcohol use Time 1 Time 2 Time 3 Marijuana use Time 1 Time 2 Time 3 Other illicit drug use Time 1 Time 2 Time 3 Dangerous driving Time 1 Time 2 Time 3

2.77 1.98 1.37 .97 1.09 1.08 2.06 2.29 2.16 .12 .14 .16 .81 .83 .64

16.59 12.06 7.13 1.80 1.86 1.89 7.99 8.17 7.99 ,139 ,135 ,159 1.58 1.38 1.04

.70 .70 .70 .70 .70 .70 .90 .90

.90
.76 .76 .76 .49 .49 .49

We evaluated the fit of Model VII as a general test of the hypothesis that there are longitudinal influences of some specific deviant behaviors on others. This model allowed for influence of each specific factor on all others at the subsequent wave. For this model to be identified, it was necessary to constrain between-behavior influences from wave one to wave two to be equal to those from wave two to wave three.6 This constraint also increases the power of the test, provided influences are roughly similar across the two time intervals. Model VII yielded a small, but significant, improvement in fit, which indicates the presence of influence between specific deviant behaviors (AVII-VI = .010, d.f. = 20, X2 = 44.01, p = .001). The coefficients for Model VII and the residuals and first derivatives for Model VI suggested that the strongest influence between specific factors was that of marijuana use on later use of other illicit drugs. In Model VIII, this path was added to those allowed in Model VI, yielding A = .986 and significant reduction in chi-square (AvIII.vI = .004, d.f. = 2, X2 =
Identification becomes an issue because there are only as many degrees of freedom arising from conelations between adjacent waves as there are possible paths between specific behaviors. Allowing all of these paths would leave no degrees of freedom for the stability of general deviance. While our solution of assuming equal influence across both intervals provides sufficient constraints to generate estimates, it is evident from the very large standard enors for all longitudinal paths that the model is empirically under-identified. Thus, Model VII is useful for assessing the potential improvement in fit from cross-behavior influences, but it does not provide meaningful estimates of those influences.

16.21, p < .001). The fit of this model was excellent, with the chi-square value virtually equal to the number of degrees of freedom. Furthermore, adding other paths between specific deviant behaviors did not significantly improve the fit of this model. Thus, we found evidence of influence between specific forms of deviant behavior, but this influence was very circumscribed.

Path Estimates Variance components. Table 2 shows the means and variances of the five measures for each wave, along with the division of the variance into general, specific, and error components that is implied by the path estimates of Model VIII.8 Our presentation of the model departs from common practice by emphasizing variance components and explained variance as much as
Though this is only one significant cross-behavior path out of a possible twenty, we are confident that it is not a chance relationship. Comparing Models VI and VII gives clear evidence of influence between specific behaviors, and the pair of paths from marijuana use to later use of other illicit drugs is significant far beyond the chance level of .05. Even allowing for the non-normality of our data, it is not plausible that a relationship of this magnitude would occur by chance. The variance components are equal to the square of a measure's loading on the relevant latent variable (lambda) times the variance of that latent variable. The variance components do not sum to the exact amount of total variance because LISREL is not constrained to precisely reproduce the diagonal of the covariance matrix. The metric of general deviance was set by fixing the lambdas for marijuana use to one.

'

GENERALITY OF DEVIANCE or more than path estimates. While this would not be desirable for most models, it is quite useful given our interest in comparing the importance of general and specific explanations for each deviant behavior. The variance components provide a straightforward comparison by combining the variance of the latent variables with their loadings on the observed variables, placing general and specific components in a shared metric. Means and variances were relatively constant over time for heavy drinking, marijuana use, and other illicit drug use, but both statistics declined for criminal behavior and dangerous driving. This age trend is well documented for both of these behaviors (Hirschi and Gottfredson 1983). Reliability is defined as the proportion of non-error variance for a measure. The measure of marijuana use had the highest reliability (.90), while the measure of dangerous driving had the lowest (.49). Reliability for the remaining measures ranged from .70 through .76. Estimates for the three types of substance use are consistent with earlier analyses of these measures (O'Malley, Bachman, and Johnston 1983). It is understandable that the measure of dangerous driving would have the lowest reliability, since this behavior was assessed indirectly through reports of traffic tickets and accidents. All estimates of paths from the general factors to the measured variables were highly significant (all t > 6.2), indicating that each behavior shared substantial variance with the others. Thus, these five deviant behaviors are potentially subject to some degree of shared explanation. Nevertheless, the importance of general deviance varied considerably across behaviors, and for some of the behaviors this proportion changed with time as well. Comparisons across behaviors are most straightforward in terms of the proportion of reliable variance (i.e., nonerror variance) associated with the general and specific factors. At all three times, criminal behavior was the form of deviance most closely associated with the general tendency, and dangerous driving was the behavior least associated. The proportion of reliable variance associated with the general factor ranged from 27 percent to 74 percent, but it rose above 50 percent only for criminal behavior. There was little variation across time in the proportion of reliable variance in marijuana use and dangerous driving associated with the general factor. Both criminal behavior and heavy alcohol use became less associated with the general deviance over time. For criminal behavior, this trend accompanied a decline in the amount and variance of the behavior. Heavy alcohol use remained equally prevalent but became more independent of other types of deviance as respondents reached legal drinking age. Over time, the use of illicit drugs other than marijuana became increasingly associated with other forms of deviance. Longitudinal relationships. Estimates for paths among the latent variables in Model VIII appear in Figure 2. Figure 2 is comparable to Figure 1, expanded to five deviant behaviors and limited to the latent variables. There was considerable longitudinal stability for both general and specific factors. The lowest unstandardized coefficient reflecting stability was .53, and eight of the twelve were above .75. Note that it is very unlikely that the stability of the specific factors would be due to memory effects such as reporting about the same incidents at more than one wave. The measures concerned behavior during the past year or less, and the interval between waves generally was two years. Only one pair of longitudinal paths indicated influence between behaviors rather than stability of a behavior. These were the paths from earlier marijuana use to later use of other illicit drugs. This influence was of moderate size for the time-one to time-two interval (standardized beta of .27), but it was insignificant for the time-two to time-three interval (standardized beta of .09). Table 3 expresses longitudinal influences in terms of the variance accounted for by measures at the preceding time.9 Model VIII allows a separation of explained variance into general and specific components, using the original metric of each variable. This provides a more direct comparison of general and specific contributions than do the path coefficients reflecting stability. The specific variance explained is a function of the amount of specific variance at the previous wave and the stability of that variance (plus any influence from other specific behaviors). The general variance explained for a behavior is a function of both the variance that behavior shares with the general factor at the current wave and the stability of the general factor. A large proportion of the reliable variance of all of the deviant behaviors can be explained by earlier measures of deviant behavior, with estimates ranging from 43 percent to 73 percent. Generally speaking, the proportion of reliable
The variance explained by specific features of a behavior equals the amount of specific variance (see Table 2) minus the unexplained specific variance (psi for that latent variable). Wave one data explained 61 percent of the variance in general deviance at wave two (the square of the stability coefficient for the standardized latent variables), and wave two data explained 60 percent of the variance in general deviance at wave three.

90

AMERICAN SOCIOLOGICAL REVIEW

54(.14),59

53(.14),60

T ~ m eOne

T~me Two

Time Three

Fig. 2. Path Estimates for the Structural Model (Model VIII): Unstandardized Path Estimates Followed by their Standard Errors (in parentheses) and Standardized Path Estimates.

variance explained by general versus specific factors is as would be expected from the breakdown o f the total variance into those two categories. Thus, most o f the explained variance for criminal behavior is due to earlier general deviance, while most o f the explained variance for marijuana use is due .to earlier specific variance. There were also some interesting variations in
Table 3. Variance Explained by Longitudinal Influences

the stability o f the specific factors. The highest proportion o f reliable variance explained was for heavy drinking at time three (71 percent) and use o f illicit drugs other than marijuana at time three (73 percent). In both cases the specific variance component was extremely stable from time two to time three (unstandardized betas o f .93 and .92). The next highest levels o f stability were for marijuana use, the other form o f

Variance Explained by Preceding Wave General

Criminal behavior Time 2 Time 3 Heavy alcohol use Time 2 Time 3 Marijuana use Time 2 Time 3 Other illicit drug use Time 2 Time 3 Dangerous driving Time 2 Time 3 2.81 1.18 .30 .21 1.51 1.5 1 ,027 ,031 .13 .09

Percent of Reliable Variance Explained Total

5.01 2.78 .73 .92 3.62 3.75 ,062 ,089 ,254 .22

Specific
2.23 1.30 .13 .68 3.08 3.21 ,035 ,055 .16 .13

General
33.5 29.7 23.1 18.6 21.6 21.3 26.3 27.7 19.3 17.0

Specific
26.5 26.1 33.1 52.3 13.3 15.7 31.3 14.9 23.7 25.7

Total
60.0 55.9 56.6 70.9 61.9 66.9 60.7 72.6 13.1 12.8

GENERALITY OF DEVIANCE substance use, and its stability did not change over time (unstandardized betas o f .83 and .84). DISCUSSION Our findings concerning general and specific features o f deviant behaviors lead to a mixed conclusion about the generality o f deviance. Hirschi (1984) and Jessor and his colleagues (Donovan and Jessor 1985; Jessor and Jessor 1977) were correct that a general tendency toward deviance could explain the positive correlations between different deviant behaviors. Indeed, a single latent variable can account for virtually all o f their cross-sectional and longitudinal relationships. Nevertheless, a latent variable o f general deviance falls far short o f explaining all o f the reliable and stable variance o f the separate behaviors. Therefore, a theory that addresses only the general construct can never fully account for the separate behaviors, though it might account for much o f each o f them. Each behavior is, in part, a manifestation o f a more general tendency and, in part, a unique phenomenon. It may still be possible to explain all o f the behaviors within a unified framework, such as each behavior resulting from peer-group norms for that particular behavior or all forms o f deviance being influenced (but not totally determined) by a general alienation. Even so, factors important to one behavior could be entirely irrelevant to others, as illustrated by Johnston's (1973) finding that support for the counterculture during the Vietnam era was strongly correlated with certain types o f illicit drug use, but not correlated with delinquency. Our analysis also provided a test for the influence o f specific deviant behaviors on one another. Marijuana use during the high school senior year had significant impact on use o f other illicit drugs one to two years later. During the subsequent two years, the influence o f marijuana use on later use o f other illicit drugs was negligible. This result suggests that any influences o f one behavior on another are age-specific, perhaps depending on age-related role transitions. While this instance o f an effect o f one behavior on another is o f interest, it is more important for our understanding o f deviance that influence o f this type was so limited. Only one o f the twenty possible paths between these five behaviors was statistically significant. Though this does not appear to be a chance relationship, it yielded a negligible improvement in the overall fit o f the model. It is clear that influences between specific behaviors contribute very little to the general finding that people who engage in one form o f deviance are likely to engage in others as well. Involvement in one form o f deviant behavior is predictive o f later involvement in others, not because o f mutual influences, but because each partially reflects a general tendency toward deviance. For instance, frequent drunkenness in the senior year o f high school would indicate a willingness to violate conventional standards o f behavior. Since the general tendency toward deviance is relatively stable over time, this willingness is likely to become manifest in other forms o f deviant behavior in the following years, as well as in the persistence o f heavy drinking. Our analysis o f the general and specific components o f each o f the five behaviors yielded interesting insights into their shifting associations over time. Criminal behavior proved to be the type o f deviance most closely linked to the general tendency, though the strength o f the tie declined over time, as did the rate and variance o f this behavior. Alcohol use among high school students was more strongly associated with a general willingness to flout conventional mores than was alcohol use among adults in their early twenties. The opposite was true for use o f illicit drugs other than marijuana, which became increasingly tied to general involvement in deviance. Though their relations to general deviance diverge, each o f these more serious forms o f substance abuse had unique aspects that became extremely stable during respondents' early twenties. Since deviance is defined by conventional standards for behavior, we would expect the overlap between general deviance and any particular behavior to fluctuate with variations in those standards. This is illustrated by our finding that heavy alcohol use is less related to general deviance once respondents reach the legal drinking age. Further research might apply our conceptual and analytic framework to additional tests o f this proposition. For instance, we would expect cigarette smoking and sexual activity to be highly related to general deviance during early adolescence because these behaviors are considered inappropriate at this age. As these activities become more acceptable in late adolescence and early adulthood, they should be less connected to the general syndrome o f deviance. In a similar vein, cross-cultural and temporal comparisons would provide a basis for testing whether normative standards influence the strength o f the connection between a behavior and general deviance. As with any piece o f research, there are limitations to our study that should be addressed by future work. W e offer a picture o f the generality o f deviance derived from samples o f 18- to 22-year olds, all o f whom had remained in high school until late in the senior year. Furthermore, while the five behaviors we have

AMERICAN SOCIOLOGICAL REVIEW considered are prominent for this age group, they hardly exhaust the meaning of the larger category of deviance. Though our findings indicate considerable generality and stability of deviance, it has yet to be determined whether this pattern holds across the full range of deviant behaviors. As with much of the research in this area, we have concentrated on the problem behaviors of adolescence and early adulthood. It is not at all clear that deviance among adults, particularly middle-class adults, would fit the same pattern. Is the scientist who falsifies research results also more likely to cheat on taxes, be unfaithful to his or her spouse, and get into fights at bars? And do adolescent behaviors such as petty theft, illicit drug use, and dangerous driving predict a broad range of adult deviance? There is much to learn about the generality of deviance.
tive Incapacitation, Cohort Studies, and Related Topics. " Criminology 24:2 13-33. Hays, Ron D., Keith F. Widaman, M. Robin DiMatteo, and Alan W. Stacy. 1986. "Structural-equation Models of Current Drug Use: Are Appropriate Models so Simple(x)?" Journal of Personality and Social Psychology 52: 134-44. Hirschi, Travis. 1969. Causes of Delinquency. Berkeley, CA: University of California Press. . 1984. "A Brief Commentary on Akers' 'Delinquent Behavior, Drugs, and Alcohol: What is the Relationship?' " Today's Delinquent 3:49-52. Hirschi, Travis and Michael Gottfredson. 1983. "Age and the Explanation of Crime." American Journal of Sociology 89:552-84. , 1986. "The Distinction Between Crime and Criminality." In Critique and E.rplanation: Essays in Honor of Gwynn Nettler, edited by Timothy Hartnagel and Robert Silverman. New Brunswick, NJ: Transaction. Jackson, Elton F . , Charles R. Tittle, and Mary J. Burke. 1986. "Offense-Specific Models of the Differential Association Process." Social Problems 33:335-56. Jessor, Richard and Shirley L. Jessor. 1977. Problem REFERENCES Behavior and Psychosocial Development: A LongitudiAkers, Ronald L. 1977. Deviant Behavior: A Social nal Study of Youth. New York: Academic Press. Learning Perspective. Belmont, CA: Wadsworth. Johnston, Lloyd D. 1973. Drugs and American Youth. . 1984. "Delinquent Behavior, Drugs, and Ann Arbor, MI: Institute for Social Research. Alcohol: What is the relationship?" Today's DelinJohnston, Lloyd D . , Jerald G. Bachman, and Patrick M. quent 3:19-47. O'Malley. 1986. Monitoring the Future: QuestionBachman, Jerald G. and Lloyd D. Johnston. 1978. The naire Responses from the Nation's High School Monitoring the Future Project: Design and ProceSeniors, 1985. Ann Arbor, MI: Institute for Social dures. (Occasional Paper # I . ) Ann Arbor, MI: Research. Institute for Social Research. Johnston, Lloyd D . , Patrick M. O'Malley, and Jerald G. Bachman, Jerald G., Patrick M. O'Malley, and Jerome Bachman. 1986. Drug Use Among American High Johnston. 1978. Adolescence to Adulthood: Change School Students, College Students, and Other Young and Stabilig~in the Lives of Young Men. Ann Arbor, Adults: National Trends Through 1985. Washington, MI: Institute for Social Research. DC: National Institute on Drug Abuse. Bentler, Peter M. and Douglas G. Bonett. 1980. Johnston, Lloyd D., Patrick M. O'Malley, and L.K. "Significance Tests and Goodness of Fit in the Eveland. 1978. "Drugs and Delinquency: A Search for Analysis of Covariance Structures." Psychological Causal Connections." Pp. 137-56 in Longitudinal Bulletin 88:588-606. Research on Drug Use: Empirical Findings and Billy, John O.G. and J. Richard Udry. 1985. "Patterns of Methodological Issues, edited by Denise B. Kandel. Adolescent Friendship and Effect's on Sexual BehavWashington, DC: Hemisphere. ior." Social Ps)~chologyQuarterly 48:27-41. Joreskog, Karl G. and Dag Sorbom. 1978. LISREL IV: Cloward, Richard A. and Lloyd E. Ohlin. 1960. Analysis of Linear Structural Relationships b.v the Delinquency and Opportunity: A Theory ofDelinquent Method of Maximum Likelihood. Chicago, IL: NaGangs. New York: Free Press. tional Educational Resources. Coleman, James S. 1961. The Adolescent Society. New Kandel, Denise B. 1978. "Homophily, Selection, and York: Free Press. Socialization in Adolescent Friendships." Amerlcan Donovan, John E. and Richard Jessor. 1985. "Structure Journal of Sociology 84:427-36. of Problem Behavior in Adolescence and Young Kandel, Denise B., Ronald C. Kessler, and Rebecca Z. Adulthood." Journal of Consulting and Clinical Margulies. 1978. "Antecedents of Adolescent InitiaPsychology 53:890-904. tion into Stages of Drug Use: A Developmental Elliott, Delbert S. and David Huizinga. 1984. The Analysis." Pp. 73-99 in Longitudinal Research on Relationship Behveen Delinquent Behavior and ADM Drug Use: Empirical Findings and Methodological Problems. Boulder, CO: Behavioral Research InstiIssues, edited by Denise B. Kandel. Washington, DC: tute. Hemisphere Elliott, Delbert S., David Huizinga, and Suzanne S. Kaplan, Howard B. 1975. Self-Attitudes and Deviant Ageton. 1985. Explaining Delinquency and Drug Use. Behavior. Pacific Palisades, CA: Goodyear. Beverly Hills, CA: Sage. Gibbs, Jack P. 1981. Norms, Deviance, and Social
Kessler, Ronald C , and David F. Greenberg. 1981 Linear Panel Analysis: Models of Quantitative Change. Control: Conceptual Matters. New York: Elsevier.
New York: Academic Press. Gold, Martin. 1970. Delinquent Behavior in an American
Krosnick, Jon and Charles M. Judd. 1982. "Transit~ons City. Belmont, CA: BrooksICole. in Social Influence at Adolescence: Who Induces Gottfredson, Michael and Travis Hirschi. 1986. "The Cigarette Smoking?" Developmental Psychology True Value of Lambda would Appear to be Zero: An 18:359-68. Essay on Career Criminals, Criminal Careers, Selec-

GENERALITY OF DEVIANCE
Lemert, Edwin M. 1972. Human Deviance, Social Problems and Social Control, 2nd ed. Englewood Cliffs, NJ: Prentice-Hall. O'Malley, Patrick M., Jerald G. Bachman, and Lloyd D. Johnston. 1983. "Reliabllity and Consistency in Self-Reports of Drug Use." International Journal oj the Addictions 18:805-24. Osgood, D. Wayne, Lloyd D. Johnston, Patrick M. O'Malley, and Jerald G. Bachman, 1985. Monitoring the Future: Folloit,-Up Srudy, Form 2, High School Senior Classes of 1976-1980, Waves 1-3 [MRDF]. Ann Arbor, MI: Survey Research Center, The University of Michigan [producer]. Sutherland, Edwin H. and Cressey, Donald R. 1955. Principles of Criminology, 5th ed. Philadelphia: Lippincott.

http://www.jstor.org

LINKED CITATIONS
- Page 1 of 1 -

You have printed the following article: The Generality of Deviance in Late Adolescence and Early Adulthood D. Wayne Osgood; Lloyd D. Johnston; Patrick M. O'Malley; Jerald G. Bachman American Sociological Review, Vol. 53, No. 1. (Feb., 1988), pp. 81-93.
Stable URL: http://links.jstor.org/sici?sici=0003-1224%28198802%2953%3A1%3C81%3ATGODIL%3E2.0.CO%3B2-R

This article references the following linked citations. If you are trying to access articles from an off-campus location, you may be required to first logon via your library web site to access JSTOR. Please visit your library's website or contact a librarian to learn about options for remote access to JSTOR.

References
Patterns of Adolescent Friendship and Effects on Sexual Behavior John O. G. Billy; J. Richard Udry Social Psychology Quarterly, Vol. 48, No. 1. (Mar., 1985), pp. 27-41.
Stable URL: http://links.jstor.org/sici?sici=0190-2725%28198503%2948%3A1%3C27%3APOAFAE%3E2.0.CO%3B2-0

Age and the Explanation of Crime Travis Hirschi; Michael Gottfredson The American Journal of Sociology, Vol. 89, No. 3. (Nov., 1983), pp. 552-584.
Stable URL:
http://links.jstor.org/sici?sici=0002-9602%28198311%2989%3A3%3C552%3AAATEOC%3E2.0.CO%3B2-6

Offense-Specific Models of the Differential Association Process Elton F. Jackson; Charles R. Tittle; Mary Jean Burke Social Problems, Vol. 33, No. 4. (Apr., 1986), pp. 335-356.
Stable URL:
http://links.jstor.org/sici?sici=0037-7791%28198604%2933%3A4%3C335%3AOMOTDA%3E2.0.CO%3B2-D

Homophily, Selection, and Socialization in Adolescent Friendships Denise B. Kandel The American Journal of Sociology, Vol. 84, No. 2. (Sep., 1978), pp. 427-436.
Stable URL:
http://links.jstor.org/sici?sici=0002-9602%28197809%2984%3A2%3C427%3AHSASIA%3E2.0.CO%3B2-J